Benchmark RadarDeskCraft
DeskCraft evaluates desktop GUI agents on long-horizon professional workflows and human-in-the-loop collaboration across 538 executable tasks in live Ubuntu desktop environments. Covers design, video, audio, and 3D creation software, with execution-based verification.
- Released
- 2026-06-02
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Existing desktop benchmarks simplify tasks and lack human-agent interaction. DeskCraft measures agent performance on realistic workflows and proactive collaboration, identifying gaps in long-horizon delivery and clarification.
Motivation
Real-world professional desktop workflows in specialized creative and engineering software unfold over long horizons and often require human-in-the-loop coordination, where agents proactively seek necessary information and users provide additional instructions, clarifications, feedback, or corrections as the task progresses.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.